The Q9400 is already as faster in a number of benchmarks and the Q9550 wins about all of the tests. There isn't even a need to get the Core7i to be faster then AMDs best. The Core7i's prices will not be dropped because there is nothing competing against them.

yeah, the PHII doesn't stand a chance against a Q95xx, not even to mention the i7.

I just wonder what kind of marketing gag the numbers on the new phenoms are... 920,940... kind of makes me wonder if this is just to scam potential customers by deliberately making it look like it could actually compete with an i7 920,940 etc...

Only because you´re focused on intels.

---

With each crime and every kindness we birth our future.

The 940 looks quite interesting -- especially when matched with a capable and BIOS updated motherboard. One consideration when comparing to the Intel's -- you do need to invest more with the Intel's -- extra cost motherboards (at least $100 more than the 790's for the Phenom II), and the extra cost for the DDR3 memory (about $50 more for 1Gx3 DDR3 compared to 2Gx2 DDR2 1066).

The really nice thing for the first Phenom II releases -- you can run them in existing AM2+ motherboards so it might be a nice upgrade from say a AMD 9600 or even a dual core in the right motherboard.

By the way, the heatsink packaged with the 940 is quite capable (as is the one included with the older 9600 Black edition AMD).

Then again, I suspect, as has been the case since early days, there is something of a performance penalty when crunching SETI work units -- like many (but not all) of the BOINC projects, there seems to have been some optimization which is Intel specific there.

I have some 740, 780 and 790 chipset boards -- they are candidates for the 940 Phenom II chips -- though only after BIOS updates find their way out (ASUS has released their BIOS updates already). Without the BIOS updates in place, one needs to manually set CPU voltages in the BIOS.

---

• Phenom II X4 920 - 2.8 GHz (1.8 GHz HT3)
  
• Phenom II X4 940 - 3.0 GHz (1.8 GHz HT3) (unlocked)
  

are the above speeds 3.0ghz per core or is the total speed 3.0ghz and you divide it by four to get the speed of each core
Thanks
Speedy

---

Per Core

---

Per Core

Thanks for your fast response. So dose this mean that there is a total processing of 12Ghz or dose it have potential to be 12Ghz?
Speedy

---

Per Core

Thanks for your fast response. So dose this mean that there is a total processing of 12Ghz or dose it have potential to be 12Ghz?
Speedy

Theoretically, yes. Realistically, no.

It is a general rule of thumb that when adding cores/CPUs, each additional core/CPU will only give you 50-75% more processing capacity.

On my 2x2 board, I ran a cinebench 10 test where I documented the effective increases in processing power by changing the affinity in task manager.



So with that being said, my clock speed per core is 1.8GHz, so if one process were to run full-out, it would be the equivalent of (1.8 * 3.6499) = 6.57GHz.

---

Linux laptop:
record uptime: 1511d 20h 19m (ended due to the power brick giving-up)

It is a general rule of thumb that when adding cores/CPUs, each additional core/CPU will only give you 50-75% more processing capacity.

That very much depends on the task, how it is programmed, and the architecture.

Plenty of interesting real-world stuff is programmed single-threaded, and gets no benefit worth mentioning from extra cores. (vastly less than 50-75)

The BOINC projects, at the opposite extreme, just start up another WU for each CPU available (whether a real physical core, or just the spare set of state of an extra instance in a hyperthreading implementation).

So BOINC work can rise very nearly in proportion to the number of cores available, depending on how much resource conflicts get in the way. For HT, the resource conflicts are bound to be substantial, and you should never expect turning HT on to come anywhere near doubling throughput. For real physical cores, on the other hand, the main resource in potential conflict is memory access (whether cache or RAM), and for real cases of interest many of the current multi-core implementations come rather close to providing linear performance improvement. (far better than 50-75)

I'd hazard a guess that BOINC work gives better output improvement for extra cores thrown to the task than almost any other currently interesting work running on consumer computers.

---

Yeah, that's right. Minor oversight. I knew I was talking about multi-threaded tasks that will use more than one core. Running a bunch of single-threaded tasks on multiple cores will nearly fully utilize all the available processing power.

Cinebench 10 ran at 3.64x compared to a single core, and when I do video encoding, I typically run about 3.4x on that (compare the wall-clock time with the CPU time in taskman).

That's all I was trying to say was that 4 cores doesn't mean 4x faster in most cases. You can expect above 3, but less than 4.

---

Linux laptop:
record uptime: 1511d 20h 19m (ended due to the power brick giving-up)

are the above speeds 3.0ghz per core or is the total speed 3.0ghz and you divide it by four to get the speed of each core

And

So dose this mean that there is a total processing of 12Ghz or dose it have potential to be 12Ghz?

Thanks to all that answered my questions
Speedy

---

There have been a few observations that the cost of the hard ware for a Phenom is cheaper than for a C2Q. But my reading is that the Phenom is not as power efficient, and that after 6 months of ownership the total cost of ownership swings to the Intel equivalent.

I'm not knocking the AMD, I always go for the most bang/buck, but in that I include the cost of electricity for the expected length of ownership, usually three years.

Up until about three years ago all my computers were AMD except for a dual P3 coppermine. They are all now Intels, and at the moment not likely to change, unless I need a server and then Opterons would be considered.

It is a general rule of thumb that when adding cores/CPUs, each additional core/CPU will only give you 50-75% more processing capacity.

That very much depends on the task, how it is programmed, and the architecture.

Plenty of interesting real-world stuff is programmed single-threaded, and gets no benefit worth mentioning from extra cores. (vastly less than 50-75)

The BOINC projects, at the opposite extreme, just start up another WU for each CPU available (whether a real physical core, or just the spare set of state of an extra instance in a hyperthreading implementation).

So BOINC work can rise very nearly in proportion to the number of cores available, depending on how much resource conflicts get in the way. For HT, the resource conflicts are bound to be substantial, and you should never expect turning HT on to come anywhere near doubling throughput. For real physical cores, on the other hand, the main resource in potential conflict is memory access (whether cache or RAM), and for real cases of interest many of the current multi-core implementations come rather close to providing linear performance improvement. (far better than 50-75)

I'd hazard a guess that BOINC work gives better output improvement for extra cores thrown to the task than almost any other currently interesting work running on consumer computers.

Spot on.

I'd add that architecture is quite important when pertaining to how the data is going to get out of RAM and into the processor. Ever since AMD has abandoned the "front side bus" architecture in favor of a "point to point" protocol that all Athlon 64 and newer chips use, they have scaled very well with additional cores due to the increase in available bandwidth. This has given AMD an edge in High Performance Computing, which typically focuses on multiple CPUs/cores and lots of data. With Intel finally abandoning the front side bus as well with their new Quick Path Interconnect, their CPUs should scale much better and will make Intel competitive in the HPC market (finally).

Of course, all this assumes that one is dealing with more data than will fit in the cache hierarchy, which is usually the case for most data.

---

Do you have some tech notes for that observation? The Phenom's I've run (these days either 125W 9850BE or 95W 9600BE) seem to run at relatively low temps (in the 40C range running BOINC so with high utilization). I had read that the 940, though a 125W processor, runs fairly cool with the stock heatsink (which I've read is quite decent) and so overclocking from 3GHz to 3.5Ghz with the stock cooler) is quite simple -- in part because it runs relatively cool and efficient.

The Intel i7 920 -- which would probably be the equivalent to the AMD 940 at about $60 more than the AMD 940 -- is a 130W processor -- so on its face, it wouldn't seem to be more efficient. I'd note that the stock fan/heatsink combination for the i7 doesn't get favorable responses, so figure another $25 to $30 there. I'm not comparing the AMD 940 to the i7 940 or i7 960 as those gems run something like $300 and $800 more than the AMD 940 (you really pay for performance there).

Then again, the DDR2 1066 memory (which is basically overclocked on its own and typically requires 2.0v to 2.2v) might well be less efficient than the DDR3 memory. Though you would be comparing say 2x2G modules versus either 3x1G modules or 3x2G modules, so the higher efficiency per module might be offset by the increased module count. I note that the DDR3 modules run faster but with more wait states (7 to 9 versus 5 for most DDR2 1066) -- not sure what that trade off amounts to performance wise.

Of course for heavy graphics users, the amount of power you use for a SLI graphics pair might dwarf what the CPU actually draws.

For me, where the current AMD Phenom II offerings are quite exciting is that they are 'drop in' upgrades for a large range of existing AM2+ configurations, so you not only don't need to get a new motherboard for upgrades, you don't need a full up reinstall (time does have value).

Later this year, the Phenom II CPU's which are NOT drop in compatible will come out -- they may well offer increased performance as well increased cost (replace the motherboard, go to DDR 3 memory, install the OS and applications from scratch). It is possible that costs there might match Intel.

There have been a few observations that the cost of the hard ware for a Phenom is cheaper than for a C2Q. But my reading is that the Phenom is not as power efficient, and that after 6 months of ownership the total cost of ownership swings to the Intel equivalent.

---

First observation I wouldn't compare Phonem with i7, unless you run i7 w/out HT, and AFAICT performance wise the AMD would be no match.

Just built in last few weeks AMD 940, GA MA790 mobo with 2 *2 GB DDR2 and Q9400, GA EP45T mobo and 2 * 2 GB DDR3. All other components same, even the cases. The Intel system was just over £40 (~$60) more expensive.
Ran them both for ~24 Hrs on BOINC, but used sons Einstein a/c as Project. The AMD uses slightly less power at idle but at 100% cpu crunching the Intel was nearly 30W watts less. That would be about 5KWh less/week if crunching 24/7.

Cannot compare performance on Einstein one was on S5R4 the other on S5R5, but think the Q9400 won that.

Do you have some tech notes for that observation? The Phenom's I've run (these days either 125W 9850BE or 95W 9600BE) seem to run at relatively low temps (in the 40C range running BOINC so with high utilization). I had read that the 940, though a 125W processor, runs fairly cool with the stock heatsink (which I've read is quite decent) and so overclocking from 3GHz to 3.5Ghz with the stock cooler) is quite simple -- in part because it runs relatively cool and efficient.

The Intel i7 920 -- which would probably be the equivalent to the AMD 940 at about $60 more than the AMD 940 -- is a 130W processor -- so on its face, it wouldn't seem to be more efficient. I'd note that the stock fan/heatsink combination for the i7 doesn't get favorable responses, so figure another $25 to $30 there. I'm not comparing the AMD 940 to the i7 940 or i7 960 as those gems run something like $300 and $800 more than the AMD 940 (you really pay for performance there).

Then again, the DDR2 1066 memory (which is basically overclocked on its own and typically requires 2.0v to 2.2v) might well be less efficient than the DDR3 memory. Though you would be comparing say 2x2G modules versus either 3x1G modules or 3x2G modules, so the higher efficiency per module might be offset by the increased module count. I note that the DDR3 modules run faster but with more wait states (7 to 9 versus 5 for most DDR2 1066) -- not sure what that trade off amounts to performance wise.

Of course for heavy graphics users, the amount of power you use for a SLI graphics pair might dwarf what the CPU actually draws.

For me, where the current AMD Phenom II offerings are quite exciting is that they are 'drop in' upgrades for a large range of existing AM2+ configurations, so you not only don't need to get a new motherboard for upgrades, you don't need a full up reinstall (time does have value).

Later this year, the Phenom II CPU's which are NOT drop in compatible will come out -- they may well offer increased performance as well increased cost (replace the motherboard, go to DDR 3 memory, install the OS and applications from scratch). It is possible that costs there might match Intel.

There have been a few observations that the cost of the hard ware for a Phenom is cheaper than for a C2Q. But my reading is that the Phenom is not as power efficient, and that after 6 months of ownership the total cost of ownership swings to the Intel equivalent.

OK - certainly for both SETI and Einstein crunching the Intel's are likely to outperform the AMD processors. I realize that as current iterations of both applications (including the optimized applications) work in a manner which seems more efficient with Intel CPU's at the software level.

My sense is that some other project applications might not operate that way -- not sure of this, I don't have current iteration Intel's systems to work with at the moment.

Further, I don't know if other applications (non BOINC) are also more Intel efficient. The theory is that if someone is looking at the fast quad processors they are not acquiring them purely for BOINC projects but rather primarily to be 'working' computers.

Regarding power consumption -- my sense there is that once you turn on hyperthreading, you run up the power consumption numbers as well -- don't know about the tradeoffs there. I do know that for at least some operating systems the general instruction was 'turn off hyperthreading' for stability reasons anyway.

As I noted before, one of the special attaction of the AMD 920 and even more the AMD 940, is that they both are 'drop in' upgrades for a large number of existing workstations. So in many cases, the cost to take a configuration running say an AMD 5000BE on a 740 chipset might be a very sustainable $235 these days plus less than a half hour change over to swap the CPU's. For that change, you are likely to more than double your BOINC crunching. So while the i7 market is new installs only (where it might well make a lot of sense), the Phenom II market is not only for cost competitive new installs, but also for the in place upgrades.

A year or so ago, I did some 'look forward but save money' deployments - AMD X2 4200's on 740 motherboards. For them, for that same $235 plus half hour effort cost, they could really goose up their workstation performance.

Then again, my typical workstation deployments are in the sub $1K hardware realm, so the i7 option is pretty much not something I look into.

---

Got the 940. It's ok. Nothing to write home about. No BIOS update yet for the MB so it comes up as an UNKNOWN CPU.

---

What particular motherboard?

Got the 940. It's ok. Nothing to write home about. No BIOS update yet for the MB so it comes up as an UNKNOWN CPU.

---

Using a MSI K9A2 SLI Platinum with the 790FX chip set, a Nivida 8800 GT Superclocked video card, and OCZ Reaper 1066mhz memory. I moved "up" from an AMD 9850 Phenom. I have the 1.6 version of the BIOS loaded which was the newest one on the MSI web site as of last night. BOINC 6.4.5 benches this rig at 2465 floating point MIPS per CPU and 7550 integer MIPS per CPU.

What particular motherboard?

Got the 940. It's ok. Nothing to write home about. No BIOS update yet for the MB so it comes up as an UNKNOWN CPU.

---

Are you overclokcing this combination at all? My sense is that the 940 will overclock by changing the multiplier from the default 15 to 17 without breaking a sweat and with other tweaks and 3rd party cooling can go significant beyond that. Are you matching it with 1066 memory as well?

Using a MSI K9A2 SLI Platinum with the 790FX chip set, a Nivida 8800 GT Superclocked video card, and OCZ Reaper 1066mhz memory. I moved "up" from an AMD 9850 Phenom. I have the 1.6 version of the BIOS loaded which was the newest one on the MSI web site as of last night. BOINC 6.4.5 benches this rig at 2465 floating point MIPS per CPU and 7550 integer MIPS per CPU.

What particular motherboard?

Got the 940. It's ok. Nothing to write home about. No BIOS update yet for the MB so it comes up as an UNKNOWN CPU.

---

No, I have not had time to play with it much as of yet. The mother board is not the greatest for overclocking. Not a wide range of adjustments available. I may take some time over the next week or so to see what I can do with it. I am not sure what you mean by "matching it with 1066 memory". I am using 4GB of OCZ Reaper 1066 memory if that is what you are asking. It is 4 1GB "sticks" so it is running in dual channel mode. For drives I am using a pair of 750GB Seagate Barracudas. I do not have the cash to burn for Velocaraptors right now. The new CPU was a big enough hit to the wallet. I am running Vista x64 at this time and I must say it is a bit funny to have the "fastest" Vista box in the local area. Yes, that was a dig at Vista performance.

Are you overclokcing this combination at all? My sense is that the 940 will overclock by changing the multiplier from the default 15 to 17 without breaking a sweat and with other tweaks and 3rd party cooling can go significant beyond that. Are you matching it with 1066 memory as well?

Using a MSI K9A2 SLI Platinum with the 790FX chip set, a Nivida 8800 GT Superclocked video card, and OCZ Reaper 1066mhz memory. I moved "up" from an AMD 9850 Phenom. I have the 1.6 version of the BIOS loaded which was the newest one on the MSI web site as of last night. BOINC 6.4.5 benches this rig at 2465 floating point MIPS per CPU and 7550 integer MIPS per CPU.

What particular motherboard?

Got the 940. It's ok. Nothing to write home about. No BIOS update yet for the MB so it comes up as an UNKNOWN CPU.

---

OK -- the MSI 790 chipset boards have gotten mixed reviews regarding performance and overclocking. With the right BIOS you should be able to change the multiplier (not the FSB) from the default 15 to 17 or 17.5 - in effect running the 940 from 3GHz x 4 CPU's to 3.5GHz x 4 CPU's. That's a 15% performance boost.

Regarding the 1066 memory, it is quite likely the motherboard is detecting it as 800 memory. For the 1066 performance to play, you need to make sure the memory voltage is bumped from a default of 1.8v to 2.1v, and then push the speed from 400 to 533. Timings (if they can be controlled) should be 5-5-5-18 for the OCX memory. Again, I'm not sure how much of this can be controlled by the MSI BIOS.

Regarding the performance, that definitely seems like less than the 940 should be doing -- it's barely quicker than my 9850 overclocked to 2.8GHz.

Oh, one more thing, the newer Windows (Windows 7) looks to be a bit faster than Vista. Of course with Vista, if you run the default extra stuff, you do waste a lot of cycles. I tend to run XP instead.

No, I have not had time to play with it much as of yet. The mother board is not the greatest for overclocking. Not a wide range of adjustments available. I may take some time over the next week or so to see what I can do with it. I am not sure what you mean by "matching it with 1066 memory". I am using 4GB of OCZ Reaper 1066 memory if that is what you are asking. It is 4 1GB "sticks" so it is running in dual channel mode. For drives I am using a pair of 750GB Seagate Barracudas. I do not have the cash to burn for Velocaraptors right now. The new CPU was a big enough hit to the wallet. I am running Vista x64 at this time and I must say it is a bit funny to have the "fastest" Vista box in the local area. Yes, that was a dig at Vista performance.

Are you overclokcing this combination at all? My sense is that the 940 will overclock by changing the multiplier from the default 15 to 17 without breaking a sweat and with other tweaks and 3rd party cooling can go significant beyond that. Are you matching it with 1066 memory as well?

Using a MSI K9A2 SLI Platinum with the 790FX chip set, a Nivida 8800 GT Superclocked video card, and OCZ Reaper 1066mhz memory. I moved "up" from an AMD 9850 Phenom. I have the 1.6 version of the BIOS loaded which was the newest one on the MSI web site as of last night. BOINC 6.4.5 benches this rig at 2465 floating point MIPS per CPU and 7550 integer MIPS per CPU.

What particular motherboard?

Got the 940. It's ok. Nothing to write home about. No BIOS update yet for the MB so it comes up as an UNKNOWN CPU.

---